Classifying Microlensing Events from ROME/REA

The Roman Galactic Bulge Time Domain Survey will complete our census of free-floating and bound exoplanets by detecting microlensing events from timeseries photometry. But the literature on classifying microlensing events is limited compared with other types of transients. Building on recent developments in machine learning techniques, this research describes the preparation and training of an eXtreme Gradient Boosting classifier to detect microlensing events in the recent data release from the ROME/REA Survey. We evaluate the classifier’s ability to distinguish microlensing from various other categories of stars, both variable and constant. In this note we discuss the preparation and filtering of the training data set and present the results of our preliminary model. One unusual feature of this data set is that it combines data from multiple telescopes, and discuss the impact this has on the training of classifiers.